| Summary: | Sentiment analysis is a powerful and popular methodology to analyze the huge amount of text information available to us every day and extract truly insightful knowledge of people’s opinion and emotional states. However, conventional methods which based on superficial features from text suffered a lot from poor affect recognition, as human emotion is often not expressed explicitly in text. The subtle difference between emotions makes the situation even more difficult. Suicide note is a great example due to the delicate emotion involved. Distinguishing genuine suicide note and hoax suicide note is usually seen as a task can only be done by trained medical professionals, because of its nature of complexity. In the research, we propose to utilize human subject self-report of valence and arousal (VA) level together with autonomic physiological signals – EEG activity to identify genuine suicide notes from other text categories including hoax suicide notes. With the help of 16 subjects who provided the VA rating and the EEG activity data while they were reading text notes, 97.22% accuracy was achieved by self-report VA ratings from the group of subjects which were provided with the class information; 83.33% accuracy was obtained from the group which was not. In the meanwhile, the model trained by EEG data from the primed group also achieved 77.3% accuracy. The findings proved the positive effect priming has on subjects and also shows the strong link between subjective self-report and objective physiological signals
|
|---|